The idea with ethereum is that oracles will be able to have a much finer granularity; whether this allows for more useful apps still remains to be proven by us in the ethereum community.
(Also, there are plenty of folks like myself working on oracle-free apps, but since these are based on unproven economic models at this point, it would be negligent for us to promote them widely before we have evidence that such designs work in practice, on the live ethereum blockchain)
"Decentralizing the web" comes from bringing in other p2p protocols for messaging and storage. Eg., if you can use cryptocurrency contracts to pay for verified storage, you can make sure files stay available without using servers.
Oracles could be useful for contracts that depend on external data but aren't the only solution.
I've still not met anybody who can tell me what Ethereum is without using the words "turing complete" (which it's not due the halting problem, just like Bitcoin). Most of the information you can get from secondary sources seems to just mirror Bitcoin (you can do this awesome thing, but you'll need an oracle to do it..).